Spend Ledger
Contents

Photo and video AI

ruDALL-E

Cover of the ruDALL-E review with the initials of the service

ruDALL-E creates images from text descriptions, with the vendor stating that prompts may be in Russian or another language. Marktechpost also describes image generation from Russian text, while reports differ on the name used for the larger model.

This may interest readers assessing text-to-image tools, provided they distinguish reported model details from current access.

Editorial review based on public sources, without hands-on testing; facts checked on 2026-11-11.

This page contains no partner link to this service.

Name
ruDALL-E
Category
Photo and video AI
Official site
rudalle.ru · No link provided
Facts checked

Three networks shape image generation

Marktechpost and 22century describe a generation, selection, and enlargement pipeline; both sources were checked on 2026-11-11.

  • Generation: ruDALL-E creates images (Marktechpost, checked 2026-11-11).
  • Selection: ruCLIP sorts images (Marktechpost, checked 2026-11-11).
  • Enlargement: SuperResolution enhances images (Marktechpost, checked 2026-11-11).
The diagram shows image generation followed by selection and image enlargement.
The sources describe a generation, selection, and enlargement sequence. Built from facts published on marktechpost.com; checked 2026-11-11.

The smaller model has a public-use report

22century reports that a smaller ruDALL-E version was made available for public use. The same source lists stock images, vector illustrations, and materials for advertising, copywriting, and design among its uses. It describes a larger version with 12 billion parameters as planned for Sber’s cloud.

The larger-model naming differs by source: Marktechpost describes XL and XXL versions, with 1.3 billion and 12 billion parameters, respectively. Pcnews.ru instead identifies Kandinsky as the 12-billion-parameter model and compares it with ruDALL-E XL at 1.3 billion parameters. Pcnews.ru says Kandinsky was trained in two phases, with the second using 119 million data pairs. It reports FID scores of 15.4 for Kandinsky and 18.6 for ruDALL-E XL.

Ai-news.ru’s example loads the pretrained Malevich model in fp16 mode on a selected device. It generates 24 images per prompt with top_k=1024 and top_p=0.99, then uses ruCLIP to select 24 from the generated set. The example applies an x4 RealESRGAN model for super-resolution.

The diagram shows a reported example from model loading through image selection to enlargement.
The example described by ai-news.ru loads a model, selects generated images, and applies super-resolution. Built from facts published on ai-news.ru; checked 2026-11-11.

Output variation complicates prompt expectations

Marktechpost reports that generation parameter ranges can differ dramatically across outputs. That is a documented source of variation, but it does not establish image quality, generation speed, or reliability. The practical details that remain open are listed under What we could not confirm.

Image selection and enlargement are included

  • 22century reports a smaller version for public use and names stock images, vector illustrations, and advertising, copywriting, and design materials among its uses.
  • A commenter on Habr reported that upscaling from the ruDALL-E library worked well; this is one comment, not broad user evidence.

What we could not confirm

  • Pricing: plans, trial, and refund terms.
  • Rules: terms of use and restrictions.
  • Company details: legal entity, address, and history.
  • Support: channels and response times.
  • Integrations: setup requirements.
  • Independent-source search: third-party sources beyond the pages gathered.
  • Blocked page: contents of the page disallowed by robots.txt.
  • Access and data handling: current service access, prompt privacy, and whether the smaller version remains accessible.
  • Larger model and technical use: later cloud availability, current access, hardware, and setup beyond the cited example.
  • Image use and product documentation: commercial rights, export formats, API access, and current documentation.

Questions about Kandinsky and ruDALL-E

What names do reports use for the larger model?

Marktechpost describes XL and XXL versions, while pcnews.ru calls Kandinsky the 12-billion-parameter model and compares it with ruDALL-E XL. The reports therefore use different naming descriptions.

What does the cited code example do after loading its model?

After loading pretrained Malevich in fp16 mode on a selected device, ai-news.ru’s example generates 24 images per prompt with top_k=1024 and top_p=0.99. It then selects 24 images with ruCLIP and applies x4 RealESRGAN for super-resolution.

Which encoding and tokenization components does the Kandinsky report name?

Pcnews.ru says Kandinsky uses Sber-VQ-GAN for image encoding and decoding, and YTTM for text tokenization. These are the components named in that report.

Fit depends on the model you can access

Fits if… the documented text-to-image scope matches your needs and you can verify which version is accessible; reports describe the smaller version differently from the larger one.

Probably not if… you need predictable generation parameters, since Marktechpost reports that parameter ranges can differ dramatically across outputs.

According to 22century.ru, a smaller version of ruDALL-E was made available for public use.

Sources

Below are the public pages behind the facts above, together with the date each one was read.

How this page was prepared

Editorial team of Spend Ledger. The editors at Spend Ledger wrote this page from the public sources listed above, which were read on 2026-11-11. No hands-on trial of the service took place. If a source changes, the page is revised.

Last updated:

Right of reply

To dispute this entry, ruDALL-E or its representatives should write to us with evidence that it is inaccurate or outdated. A documented refutation leads to a correction or removal of the entry. No payment changes what we publish.

Our contact for replies
No data available
Response time
No data available
Back to the reviews